1
Wage Functions and Rates of Return to Education in
Italy
Davide Fiaschi and Cecilia Gabbriellini1
University of Pisa
June 17, 2013
Abstract
We study the return to education in Italy in the period 1995-2010 for a
representative sample of Italian households. In line with previous
literature, OLS under-estimate the return to schooling. When the
endogeneity of schooling is taken into account, the return to an
additional year in school increases. The evidence is that returns have not
changed much over the considered period, varying between 5.9% and
7.9%. Looking to the different sector of employment, a relative
convenience to work in the public sector emerges, but not significant for
all the analyzed years. In addition, there is an evidence of a gender pay
gap, in favor of men for all the period considered. When the type of
school attended is taken into consideration, the returns to education
increase with higher levels of educational attainment.
Keywords: education premium, Mincer equation, gender gap, public sector
JEL: I21; J24; J31; J45; J71
1 Corresponding author: Cecilia Gabbriellini, Dipartimento di Economia e Management, University of Pisa, Via
Ridolfi 10, 56124, Pisa (PI), Italy, e-mail: [email protected].
2
1 Introduction
Education is one of the most important components of individual human capital
(Becker 1993) thus a significant determinant of earnings. The estimation of the
economic return to education has perhaps been one of the predominant areas of
analysis in applied economics for over 50 years, in both micro and
macroeconomics. The analysis of education has been driven by the concept of
human capital, pioneered by Gary Becker, Jacob Mincer and Theodore Schultz. In
human capital theory, education is an investment of current resources for future
returns. The benchmark model for the development of empirical estimation of the
returns to education is the relationship derived by Mincer (1974) between log
hourly earnings and schooling. The original Mincer equation assumes linear effect
on earnings of each year of education regardless of the attainment level.
The aim of this paper is to evaluate returns to education in Italy over time. In
particular, we focus on education as a private decision to invest in human capital
and we explore the internal rate of return to that private investment. Then, we take
into account differential effects of different educational level (upper-secondary and
tertiary education) and of different type of school (scientific, humanistic, technical)
within each educational level. We use the SHIW conducted by the Bank of Italy,
covering the period from 1995 to 2010 to estimate the returns to education.
The reminder of the paper is organized as follows. Section 2 provides some
descriptive statistics on educational attainments in OECD Countries. Section 3
reviews previous works in the area. Section 4 presents the theoretical background
of the earnings equation to be estimated. Section 5 describes the dataset used in the
empirical estimation and the characteristics of the sample. Section 6 reports the
estimates of the effect of schooling on individual wages. Finally, section 7
summarizes and concludes.
3
2 Education Attainments and Education Premium in OECD Countries
The current Italian education system is composed by primary, secondary, upper
secondary and tertiary education. Primary school is compulsory for children aged
between 6 and 11 years. Lower secondary education is also compulsory, free of
charge and lasts three years. Post compulsory education is differentiated into the
following categories: classical, scientific and pre-school teacher training, artistic
education, technical school and vocational education. Upper secondary education
lasts from three to five years, depending on the type of school. Since 1969, the
selection of the type school does not preclude access to tertiary education.
Graduation from upper secondary schools requires a leaving school certificate
examination and access to tertiary education is only conditional on passing this
exam.
In comparison with other OECD countries in 2009, average education attainments
of the upper secondary education in Italy is substantially low as shown in Table 1.
On average across OECD countries, the percentage of 25-34 year-olds with at least
upper secondary education is 20 per cent higher than that among 55-64 year-olds
(about 81.5 per cent against 61.3 per cent). This difference for cohort can be
explained by the observed general decline in demand for manual labor and for basic
cognitive skills (easily replicated by computers), in favor of a sharp increase in the
demand for complex communication and advanced analytical skills, which require
a more educated labor force.
Table 1 - Percentage of population that has attained at least upper secondary education, by age
group.
25-34 years old 55-64 years old
OECD average 81.5 61.3
Italy 70.3 36.7
Source: OECD (2011)
In Italy, just 70.3 per cent of the age-group 25-34 (versus an OECD average of 81.5
per cent) has attained at least upper secondary education; however, such a
4
percentage is much higher than the 36.7 per cent of the 55-64 age-group. Indeed, in
Italy since 2000 to 2009 the percentage of population with upper secondary
education increased by an average of 3 per cent, which represents an annual growth
rate of 0.4 per cent. Extrapolating the current patterns of graduation, an average of
81 per cent of today’s young people will complete upper secondary education over
their lifetimes.
As regards tertiary education in OECD countries we observe the same upward trend
of education attainment for younger cohorts of population as reported in Table 2
(from 22.4 per cent to 37 per cent).
Table 2 - Percentage of population that has attained tertiary education, by age group.
25-34 years old 55-64 years old
OECD average 37.0 22.4
Italy 20.2 10.3
Source: OECD (2011)
In Italy in 2009 the percentage of population in the 25-34 years-olds cohort with a
university degree is equal to 20.2 per cent, much lower than the OECD average of
37 per cent. Even though Italy shows a very significant increase over time of the
percentage of the population attaining tertiary education (20.2 per cent of the 25-34
age group must be compared with 10.3 per cent of the 55-64 age group), we notice
that such difference is well below that observed for OECD countries (from 22.4 per
cent to 37 per cent).
Looking at gender in OECD and Italy, evident disparities in educational
attainments between women and men are present in the older generations, but with
a significant inversion in the more recent cohorts (see Tables 3 and 4). In particular,
in OECD countries while for older generation (e.g. 55-64 age group) the percentage
of people attaining upper secondary and tertiary education is significantly larger for
men, for the 25-34 age group the educational level is significantly higher for
women.
5
Table 3 - Percentage of population that has attained at least upper secondary education, by gender.
Women, by age group
25-64 25-34 35-44 45-54 55-64
OECD average 72 83 77 69 57
Italy 55 74 61 51 33
Men, by age group
25-64 25-34 35-44 45-54 55-64
OECD average 74 80 77 72 66
Italy 54 67 55 50 41
Source: OECD (2011)
The gender gap in education in favor of women is recorded also in Italy: 7 per cent
higher for the same group for upper secondary education, and 9 per cent higher for
women aged 25-34 for tertiary education.
Table 4 - Percentage of population that has attained tertiary education, by gender.
Women, by age group
25-64 25-34 35-44 45-54 55-64
OECD average 31 42 34 27 21
Italy 16 25 17 12 10
Men, by age group
25-64 25-34 35-44 45-54 55-64
OECD average 29 33 30 27 24
Italy 13 16 13 12 11
Source: OECD (2011)
Tertiary education brings substantial economic benefits for workers both in terms
of higher earnings and lower probability to be fired. A person with a tertiary
education can expect to earn over 50 per cent more than a person with an upper
secondary or postsecondary non-tertiary education (OECD, 2011). In OECD
countries, those who do not complete an upper secondary education could earn an
average of 23 per cent less than their counterparts who do complete that level of
education. The earnings advantage of having a tertiary degree increases with age
(on average, the earnings of tertiary-educated 55-64 year-olds is larger than that for
6
25-64 year-olds: by 13 per cent for OECD countries, by 45 per cent for Italy) but
across all educational levels women earn considerably less than men (in Italy,
women who have obtained a tertiary degree earn 65 per cent or less of tertiary-
educated men).
In all OECD countries, individuals with a tertiary-level degree have a greater
chance of being employed than those without such a degree. In general, higher
education improves job prospects and the likelihood of remaining employed in
times. In 2009, in Italy 79 per cent of the population with a tertiary education is
employed against 73 per cent with a upper secondary education (84 per cent against
74 per cent in OECD countries). Employment rates for workers with a tertiary
education are higher of 28 per cent with respect to workers who have not completed
an upper secondary education either for OECD countries and for Italy.
Finally, also the effect on earnings of an upper secondary education changed over
time. Young individuals (25-34 year-olds) with a vocational2 upper secondary
education typically do well in the labor market when compared with the total 25-64
year-old population. In Italy, the unemployment rate for young individuals (25-34
year-olds) with vocational upper secondary education is 3 per cent points higher
than that of the 25-64 year-old population (OECD, 2011).
3 Literature on the Estimated Returns to Schooling in Italy
The main features of empirical research on returns to education in Italy are shown
in Table 5. The estimated rate of return to an additional year of schooling
considerably vary across studies, also for the method used in the estimate. Antonelli
(1985), who consider regional data, estimates that an additional year of schooling
increases annual net earnings by 4.6 per cent. Cannari et al. (1989) use a larger
sample from the 1986 wave of the Bank of Italy, finding a similar result of a return
around 4 per cent. While Lucifora and Reilly (1990) estimate the mincerian
earnings function using the ENI special survey on earning and they find that the
2 Vocational or technical education is defined as education that is mainly designed to offer participants the opportunity to acquire
7
marginal return to schooling is slightly higher for men than for women but again
around 4 per cent.
Table 5 – A summary of the estimated return to schooling of an additional year of schooling in Italy
of a sample of empirical contributions.
Author Method Years of observations Marginal return to
education
Antonelli (1985) OLS 1977 4.6
Cannari, Pellegrini, and Sestito (1989) OLS 1986 4.0
Lucifora and Reilly (1990) OLS 1985 4.0 (men) 3.6 (women)
Cannari and D'Alessio (1995) IV 1993 7.0
Colussi (1996) IV 1993 7.6
Flabbi (1997) IV 1991 6.2 (men) 5.6 (women)
Brunello and Miniaci (1999) IV 1993 and 1995 5.7
Brunello, Comi, and Lucifera (2000) OLS 1995 6.2 (men) 7.7 (women)
Ciccone (2004) OLS 1987-2000 6.1
Ciccone, Cingano, and Cipollone (2006) OLS 1987-2000 6.9
Mendolicchio (2006) PV 2002 5.3 (men) 6.5 (women)
Cingano and Cipollone (2009) OLS 1987-2000 6.0
For the 1993 wave of Bank of Italy Cannari and D’Alessio (1995), using family
background variables as instruments of educational outcomes, find that the
marginal return to education is around 7 per cent, much higher than previous
results. Also Colussi (1996) obtain a similar result, using the same wave and a
similar set of instruments. For 1991 wave Flabbi (1997) calculates the returns to
education separately for men and women with an instrumental variable approach
based upon the identification of exogenous changes in the schooling system; he
finds that the marginal effect of education is 6.2 per cent for men and 5.6 per cent
for women, confirming the gender gap in earnings. For the 1993 and 1995 waves,
Brunello and Miniaci (1999) estimate a return to education equal to 5.7 per cent
(taking into account the endogeneity of schooling). The estimated coefficient on the
mincerian rate of return to schooling is around 6 per cent in Ciccone (2004) and
Cingano and Cipollone (2009).
Brunello, Comi and Lucifora (2000) find evidence of a greater return to schooling
for women, that is also confirmed in the work of Mendolicchio (2006), in which
proxy variables approach is applied to deal with the endogeneity of the schooling
variable.
8
In summary, in the analysis the check for the endogeneity appears a crucial feature
in order to avoid a likely downward bias in the estimate; furthermore, the estimated
return appears to be varying over time, suggesting to separately consider the
different waves; and, finally, gender gap should be take into consideration.
4 Theoretical Background Model and Empirical Strategy
The theoretical approach underlying most empirical studies of schooling attainment
is the model of accumulation of human capital developed by Schultz (1961),
Becker (1964) and Mincer (1958, 1974). In human capital theory, education is an
investment of current resources in exchange of future returns. The benchmark
model for the empirical estimation of the returns to education is the relationship
developed by Mincer in a very celebrate model in 1974. This model focuses on the
life-cycle dynamics of earnings and on the relationship between observed earnings,
potential earnings and human capital investment, both in terms of formal schooling
and on-the-job investment. No explicit assumption are made about the background
economic environment. Observed earnings are a function of potential earnings net
of human capital investment costs, where potential earnings in any time period
depend on investments made in previous time periods. Let be the potential
earnings at time t. Investments in training can be expressed as a fraction of
potential earnings invested, i.e. , where is the fraction invested at time
t. Let be the return to training investments made at time t. Then:
(1)
Repeated substitution yields .
Formal schooling is defined as years spent in full-time investment ( .
Assume that the rate of return on formal schooling is constant for all years of
schooling ( ) and that formal schooling takes place at the beginning of life.
Then, assume that the rate of return to post-school investment, is constant over
time and equals . Then, we can write:
9
(2)
which yields the approximate relationship (for small and )
(3)
To establish a relationship between potential earnings and years of labor market
experience, Mincer (1974) approximates the Ben-Porath (1967) model and further
assumes a linearly declining rate of post-school investment:
(4)
where is the amount of work experience as of age t. The length of
working life, , is assumed to be independent of years of schooling. Under these
assumptions, the relationship between potential earnings, schooling and experience
is given by:
(5)
Observed earnings equal potential earnings less investment costs, producing the
following relationship for observed earnings:
(6)
Then, this is the standard form of the Mincer earnings model that regresses log
earnings on a constant term, a linear term in years of schooling, and linear and
quadratic term in years of labor market experience. In most of applications of the
Mincer model, it is assumed that the intercept and slope coefficients are identical
across persons. This implicitly assumes that , , and are the same across
persons and do not depend on the schooling level. However, Mincer formulates a
10
more general model that allows for the possibility that and differ across
persons, which produces a random coefficient model:
(7)
Assuming , we can write this
expression as:
(8)
where the terms in brackets are part of the error. Initially, Mincer assumes that
are independent of In later
work (Mincer, 1977), he relaxes this assumption.
Mincer derives several implications from the accounting identity model under
different assumptions about the relationship between formal schooling and post-
school investment patterns. Under the assumption that post-school investment
patterns are identical across persons and do not depend on the schooling level, he
shows that
and
. These two conditions imply:
(i) log-earnings experience profiles are parallel across schooling levels
(ii) log-earnings age profile diverge with age across schooling levels.
In the Mincer specification, the disturbance term captures unobservable individual
effects, as unobserved ability, that may also influence schooling decision, and then
induce a correlation between schooling and the error term in the earnings function.
The correlation between schooling and error term means the former does not
measure casual effects and thus schooling is endogenous. With endogeneity the
returns to schooling estimated by the then estimation by ordinary least squares are
biased. The solution to this problem has been addressed in different ways. Firstly,
the measures of ability have been incorporated with a proxy variable for
unobserved effects, in order to control separately the effect of education and ability.
Secondly, one might exploit within-twins differences in wages and education,
11
assuming that unobserved effects are additive and common within twins so that
they can be differentiated out by regressing the wage difference within twins
against their education differences. Another approach deals with the simultaneous
relationship between schooling and earnings by specifying a two-equation system,
which is identified by exploiting instrumental variables that affect s but not w.
Instrumental variables estimation, using family background as instruments for
schooling, will be our strategy to deal with endogeneity.
The last approach is the most applied in the literature and it is what we applied in
the following estimates.
5 The Dataset
The analysis is based on data provided by the Bank of Italy’s Survey of Household
Income and Wealth (SHIW), reporting several socio-economic characteristics of
Italian households. The SHIW is a biannual survey on the microeconomic behavior
of Italian families with a sample of approximately 8,000 household per year. The
observations from eight subsequent surveys will be analyzed, from 1995 to 2010. In
particular, the SHIW contains information both on households (family
composition) and on individuals. Moreover, it provides detailed information on
several characteristics of the worker, such as net yearly earnings, average weekly
hours of work and number of months of employment per year3, educational
attainment (the highest completed school degree4), job experience, gender, marital
status, sector of employment, composition of his/her family, parents background,
regions of residence, town size.
3 Our definition of the hourly wages is: yearly net earnings/(months worked*weekly hours worked*4)
4 Standard and not actual year of formal schooling are recorded. Since students who fail to reach a standard have to
repeat the year, the actual number of years is likely to be underestimated.
12
We consider a sub-sample of men and women between 15-64 years old, full time
and part time employees, working either in the public or in the private sector5 and
such that information about earnings are available.
5.1 Variables Used in the Analysis
Earnings, schooling attainment, and working experience of every individual are the
key variables in the estimate of Mincer equation.
In its original formulation, the Mincer equation refers to the hourly price of labor as
correct measure of worker’s earnings6. SHIW contains annual earnings net of taxes
and social security contributions. Additional information on the average number of
hours worked per week and on the number of months worked per year can be used
to construct an estimate of the hourly net wage, the measure used by most empirical
studies7.
Schooling attainment is generally measured by the number of years spent at school.
SHIW does not contain information about this number of years, but only on the
highest degree attained by individual. Following a common approach in literature
(see, for instance, Vieira, 1999; Brunello and Miniaci, 1999) we calculate the
educational attainment of the individual by imputing the number of years required
to complete her/his reported level of educational attainment8. More precisely, we
consider that the (statutory) numbers of years required to obtain a primary and a
junior school certificate is 5 and 8 years respectively; instead, for the upper
secondary school the number of years ranges from 11 (vocational or technical
school) to 13 (classical or scientific studies); finally, for tertiary education, we
5 We exclude self-employed because of the low reliability of their declared earnings. As calculated by Brandolini and
Cannari (1994) Bank of Italy’s Survey of Household Income and Wealth (SHIW) seems to underestimate the self-
employed earnings of about 50 percentage points.
6 Monthly or annual wages would in addition capture the effect of individual’s decisions on working hours. Given
the only weak positive correlation between working time and educational attainment it is reasonable to assume that
the choice of hours worked reflects individual preferences rather than educational levels.
7 Notice that hourly measure of earnings can be affected by measurement errors due to the fact that we calculate
hourly wages as total earnings divided by hours of work.
8 Standard, not actual, years of formal schooling are recorded. Since students who fail to reach a standard have to
repeat the year, the actual number of years is likely to be underestimated.
13
consider 16, 18 and 21 years for the university diploma, the college degree, and the
postgraduate degree (e.g. Ph.D.) respectively. In the analysis we will also treat
education as a categorical variable divided into 6 categories: no education, primary
school, junior high school, 3-year vocational school, upper secondary school,
tertiary education (including university diploma, college and post-graduate
education). It is important to stress that the statutory number of years can be
significantly different from the actual number of years spent to obtain a degree,
especially at college because of the high percentage of irregular student.
Many empirical studies use age as a proxy for the (working) experience of
individuals. But this choice can be severely biased, especially for young cohorts.
Other authors use potential experience, defined as the difference between the
current age and the age at the labor market entry, but they ignore the possibility of
unemployment or underemployment, again a crucial feature for young cohorts.
Here we use as proxy for experience the number of years for which a worker has
been paid social security contribution; they should reflect the effective years of
training on the job and learning-by-doing activities.
The control variables are as follows. A gender dummy (DUMMY_MALE) controls
for different wage levels between men and women. Marital status also enter into the
analysis as a dummy variable (DUMMY_MARRIED) taking the value 1 if the
person is formally married and 0 otherwise. Over and above its effect via hours
worked, part-time work is captured through a separate dummy variable
(DUMMY_PART_TIME) since the assumption that each working hour makes the
same contribution to weekly earnings (constancy of the hourly wage) may not hold
across workers with different time status (part time versus full time).
We also introduce controls for family composition, as a proxy for the influence of
housework, particularly important in the female labor supply (Heckman and
Killingsworth, 1986). Indeed we control for the number of components of the
family (NCOMP) and moreover, we add a dummy (DUMMY_HOUSEHOLD) to
take into account the fact that the individual is the household of the family.
14
Controls for sector in the labor market (DUMMY_AGRICULTURAL for the
agricultural sector, DUMMY_INDUSTRIAL for the industrial sector,
DUMMY_PUBLIC for the public sector and D_OTHER_SECTOR for other sector
different from the previous ones) concern the demand side factors that are not fully
explained by human capital variables, and they allow reducing the bias arising from
an imperfectly competitive labor market. Finally, the regression controls for the
geographical area of residence: one dummy for the town of residence that has more
than 500.000 inhabitants (DUMMY_TOWN), and three different dummies for the
macro-regions: Nord, Center and South (DUMMY_NORTH, DUMMY_CENTER
and DUMMY_SOUTH).
Table 6 reports the descriptive statistics of the main variables used in the empirical
analysis for all the waves. Notice that the available measure of earnings is net of
income taxes and social security contributions.
Table 6 - Means and standard deviations of the variables used in the empirical analysis
for the entire sample (1995, 1998, 2000, 2002, 2004, 2006, 2008, 2010)
Variable Mean Stand. Dev. Description
LOGY_H 2.210 0.439 Logarithm of the hourly real earnings less tax
SCHOOL 11.294 3.796 Schooling attainment, that is the number of years
spent at school
COMP_SCHOOL 0.391 0.488 Compulsory school: no schooling, primary school
and junior high school
VOCATIONAL 0.091 0.287 3-years Vocational degree
UPPER_SECONDARY 0.377 0.485 Upper secondary degree
DIPL_TECN 0.212 0.409 Technical school
DIPL_LIC 0.056 0.230 Liceo classico, scientifico, linguistico, artistico
DIPL_PROF 0.050 0.219 Vocational school
DIPL_MAG 0.050 0.217 Liceo magistrale
DIPL_OTHER 0.008 0.089 Other upper secondary degree
TERTIARY 0.141 0.348 Tertiary degree
L_LETT 0.039 0.195 Literature, Philosophy, Languages, Psychology
L_SCIEN 0.047 0.212 Mathematics, Physics, Chemistry, Biology,
Medicine, Engineering, Agriculture, Veterinary
L_UMAN 0.038 0.190 Economics, Statistics, Architecture, Political
Science, Sociology, Law
L_OTHER 0.017 0.128 Other tertiary degree
EXPERIENCE 17.480 10.616
Number of years for which it has been paid social
security contributions, as a proxy for years of
training on the job
DUMMY_MALE 0.581 0.493 Gender dummy
DUMMY_MARRIED 0.647 0.478 Dummy variable for marital status
15
NCOMP 3.343 1.179 Number of components of the family
DUMMY_HOUSEHOLD 0.473 0.499 Household dummy, that is equal to 1 if the
individual is the household of the family
DUMMY_PART_TIME 0.087 0.281 Dummy variable for part time work
DUMMY_AGRICULTURAL 0.034 0.180 Dummy variable for agricultural sector
DUMMY_INDUSTRIAL 0.324 0.468 Dummy variable for industrial sector
DUMMY_PUBLIC 0.324 0.468 Dummy variable for public administration sector
DUMMY_OTHER_SECTOR 0.318 0.466 Dummy variable for other sector
DUMMY_TOWN 0.083 0.276 Dummy variable for the town of residence that has
more than 500.000 inhabitants
DUMMY_NORTH 0.503 0.500 Dummy variable for North regions
DUMMY_CENTER 0.213 0.409 Dummy variable for Center regions
DUMMY_SOUTH 0.284 0.451 Dummy variable for South regions
DUMMY_SECT_PARENTS 0.379 0.485
Dummy variable equal to 1 if the individual works
in the same sector of the father and/or of the
mother
SCHOOL_F 6.040 4.095 Schooling attainment of the father's worker
SCHOOL_M 5.284 3.693 Schooling attainment of the mother's worker
6 Results
6.1 Estimation of Schooling
For each available year, a cross-sectional OLS regression of the Mincerian wage
equation described in (6) is run to stress whether the estimated returns to education
have varied significantly over time. Robust standard errors are computed, in order
to control for the presence of outliers and heteroskedasticity (see the results in the
appendix).
However, as discussed in a very large literature summarized by Card (1994),
ordinary least squares estimates of the returns to education are not consistent either
because of measurement errors in the schooling variable or because of the
endogeneity of the schooling variable.
In particular, the measurement of years of schooling in our data is exposed to error
because we lack information on completed years and observe only the last
completed degree. But individuals with the same completed degree could have
spent a significantly different number of years in education. Moreover, the
endogeneity bias may arise either from unobserved variation in ability or from
unobserved heterogeneity. If those who extended education beyond compulsory
schooling have greater ability than other, then the estimated return to education is
16
biased upwards since part of the productivity differential is due to ability or skills
acquired outside the school (ability bias). Thus, the ability bias may interact with
heterogeneous subjective discount rates that result in under-estimating the true
effect of schooling on earnings if the more impatience individuals happen to be the
more able ones (heterogeneity bias). The total effect of the bias in the OLS
estimates is ambiguous.
One way to deal with measurement errors and the endogeneity of schooling is to
estimate the eq.(6) by using instrumental variables (IVs). The identification of a
valid instrument is not easy work and has been reviewed among others by Card
(1999) and Ashenfelter, Harmon and Oosterbeek (1999).
The requirements for an instruments to be valid are that it should be correlated with
educational choice but not correlated with log wages conditional on schooling
(Wooldridge, 2002). Candidates to be used as an instrument are only weakly
correlated with the endogenous variable in question. It is well recognized that using
such variables as instruments is likely to produce estimates with large standard
errors. In particular, if the correlation between the instrument and the endogenous
explanatory variable is weak, then even a small correlation between the instrument
and the error can produce a larger inconsistency in the IV estimate of the
coefficients than in the OLS estimates (Bound et al., 1995).
There is a long tradition in using family background variables, typically the level’s
of parent’s schooling, as a valid instruments (Cannari and D’Alessio, 1995;
Colussi, 1997; Card, 1999). The idea is based on the observation of persistence
across generation about the level of schooling and it is theoretically justified by
involuntary transmission of human capital.
Our instruments will be a set of variables that measure family background,
including the highest completed educational level by the father and the mother of
the interviewed individual. Then, more educated parents are likely to value
education more and to fill better jobs.
However, the selection of family background variables as additional instruments
has two potential problem. First, individual is asked to recall both the highest
17
educational level and the occupation held by his parents when they had his current
age. Then, it is not clear whether information based on the same age as the
respondent is always the most relevant. This is the case especially for the
profession of the parents, that could have changed with respect to the profession
held during the schooling period of the interviewed individual. Second, family
characteristics could affect the returns to education, thus failing to satisfy the
necessary condition for instruments validity. In our estimation strategy, the
instrument validity are tested by computing Sargan test, which is an over-
identification test with an asymptotic χ² distribution and degrees of freedom equal
to the number of over-identifying restrictions. The test verifies whether the
instruments play a direct role, through predicting educational attainment
(Wooldridge, 2002). An important requirement in also that selected instrument
should be correlated with the endogenous variable. The excluded instruments in the
reduced form schooling equation are tested by computing the F-statistic on the
excluded instruments in the reduced form schooling equation, experience and the
square of experience equations as suggested by Bound et al. (1995).
Table 7 - IV estimates9. Dependent Variable: log of hourly earnings less tax. Omitted categories are:
Center (DUMMY_CENTER); Industrial sector (DUMMY_INDUSTRIAL).
VARIABLES 1995 199810 2000 2002 2004 2006 2008 2010
SCHOOL 0.0643*** 0.0619*** 0.0686*** 0.0712*** 0.0668*** 0.0786*** 0.0587*** 0.0685***
(0.00368) (0.00764) (0.00477) (0.00759) (0.00678) (0.00621) (0.00613) (0.00784) EXPERIENCE 0.0189*** 0.0188*** 0.0206*** 0.0246*** 0.0144*** 0.0250*** 0.0227*** 0.0151***
(0.00331) (0.00671) (0.00354) (0.00530) (0.00450) (0.00375) (0.00447) (0.00439)
EXPERIENCE^2 -0.00014* -0.00014 -0.00019** -0.000278** -0.000149 -0.000326*** -0.000271** -4.20e-05 (7.92e-05) (0.000155) (8.13e-05) (0.000128) (0.000115) (9.04e-05) (0.000112) (0.000100)
DUMMY_MALE 0.132*** 0.114*** 0.0940*** 0.0983*** 0.0812*** 0.109*** 0.157*** 0.154*** (0.0250) (0.0354) (0.0187) (0.0252) (0.0261) (0.0211) (0.0215) (0.0266)
DUMMY_MARRIED 0.00438 0.0501 0.0555** 0.00825 0.0369 -0.00941 -0.0500* 0.0292
(0.0249) (0.0448) (0.0251) (0.0381) (0.0317) (0.0235) (0.0279) (0.0283) NCOMP 0.0177** 0.0150 -0.000804 0.00124 -0.00221 0.0315*** 0.0275*** -0.00232
(0.00728) (0.0146) (0.00781) (0.0101) (0.00898) (0.00896) (0.00893) (0.0110)
DUMMY_HOUSEHOLD -0.00637 -0.00119 0.00604 0.0224 0.0188 0.0306 0.00901 (0.0254) (0.0369) (0.0184) (0.0247) (0.0245) (0.0200) (0.0237)
DUMMY_TOWN 0.00582 0.0310 0.0128 -0.0815** -0.0184 0.0423* 0.0166 -0.0339
(0.0210) (0.0405) (0.0215) (0.0360) (0.0447) (0.0242) (0.0305) (0.0395) DUMMY_NORTH 0.0378** 0.0671** 0.0464*** 0.0456* 0.0666** -0.00828 -0.00206 0.0514*
(0.0167) (0.0286) (0.0171) (0.0238) (0.0297) (0.0193) (0.0230) (0.0288)
DUMMY_SOUTH -0.0239 0.0634** -0.00599 0.00615 0.0224 -0.0493** -0.0344 0.0201 (0.0185) (0.0319) (0.0224) (0.0280) (0.0353) (0.0230) (0.0254) (0.0316)
9 In the SHIW waves, information about family background is available only for the households and for his/her
spouse or cohabitant. For year 2008 for the households and for his/her spouse or cohabitant if the households is borne
in an odd year, while for year 2010 only for the households.
10 The sample of the wave of year 1998 is smaller in comparison to other waves, because of the lack of data on the
variable experience. This fact may affect the estimation.
18
DUMMY_AGRICULTURAL -0.0394 0.0210 -0.118* -0.0405 -0.0661 -0.127* -0.0606 0.0278
(0.0703) (0.104) (0.0634) (0.0578) (0.0435) (0.0742) (0.0481) (0.0692)
DUMMY:_PUBLIC 0.109*** 0.0434 0.0191 0.00784 0.0525* 0.0100 0.0947*** 0.0677* (0.0218) (0.0343) (0.0215) (0.0314) (0.0311) (0.0290) (0.0288) (0.0356)
DUMMY_OTHER_SECTOR 0.0156 -0.00719 -0.00734 -0.0140 -0.0144 -0.0298 -0.00225 0.00405
(0.0179) (0.0397) (0.0197) (0.0232) (0.0263) (0.0204) (0.0235) (0.0265) DUMMY_SECT_PARENTS -0.00735 0.0422* -0.0132 -0.00237 -0.0113 -0.0150 0.0306* 0.00793
(0.0182) (0.0250) (0.0151) (0.0186) (0.0188) (0.0172) (0.0186) (0.0224)
DUMMY_PART_TIME 0.0387 0.0777 0.0673** -0.0617 -0.0124 0.0180 0.0186 0.0444 (0.0360) (0.0648) (0.0340) (0.0440) (0.0386) (0.0415) (0.0367) (0.0333)
Constant 1.074*** 1.028*** 1.061*** 1.033*** 1.183*** 0.923*** 1.115*** 1.026***
(0.0596) (0.132) (0.0697) (0.0940) (0.0854) (0.0879) (0.0878) (0.112)
Observations 4,352 1,468 3,783 3,321 3,405 3,437 2,836 2,145
R-squared 0.403 0.308 0.293 0.261 0.206 0.268 0.331 0.250
Robust standard errors in parentheses
*** p<0.01, ** p<0.05, * p<0.1
Table 7 presents the IV estimates for the period 1995-2010. The Sargan test never
rejects the null hypothesis of no miss specification (see the first stage estimation
and all the tests in the appendix), so we cannot reject the validity of over-
identifying restrictions. In addition the Bound test that always rejects the null
hypothesis of no correlation between education and additional instruments.
We confirm for this sample the finding that the estimated returns to education are
significantly larger with IV than with OLS, as stressed by large part of the
international literature. The downward OLS bias implied by IV estimates could
arise from the attenuation effect of a measurement error in the schooling variables,
but also a distortion from omission of the variable “ability” could lead to a similar
result. This means that the more “able” (in terms of capacity to earn higher wages)
individuals have lower preference for schooling, and those preferences could be
justified by the higher opportunity costs faced by the “able” individuals. However,
there is also another possible explanation of the higher IV estimates, in case of
heterogeneity in the economic benefit of schooling, as well as heterogeneity in
costs or tastes for schooling. Assume the existence of an exogenous event that
reduces the marginal costs of schooling only for a subsample of the population.
Therefore, it is possible to define an instrumental variable assuming value 1 on this
subsample (treatment group) and value 0 on the rest of the population (the control
group). The reason of the bias is clarified considering that an instrument that
reduces the cost of schooling is more effective on people with the higher marginal
cost of education. Then, if the cost component plays a minor role in an individual
school choice, then the individual would be virtually unaffected by an exogenous
19
event that influences costs11
. This means no random selection of the treatment
group, implying the violation of the necessary assumption for IV and lending to an
inconsistent estimation. The bias is positive if the average marginal return in the
treatment group is higher than the average marginal return in the overall
population, otherwise the bias is negative. The same idea is formalized in the Local
Average Treatment Effect literature12
: the IV estimate is the average return to
schooling for people that acquire more education only because of this particular
exogenous event and that would have not acquired additional education in the
absence of this particular event.
6.1.1 The Returns to Schooling
The evidence is that returns have not changed much over the period considered.
The estimates of the returns to schooling are between 5.9% and 7.9%, recording the
highest level for 2006. Looking at the previous estimates made for Italy, as shown
in table 5, we can notice that our estimate are in line with the literature. Moreover,
it is not present a clear patterns of the return to schooling, either increasing or
decreasing.
Table 8 - Estimates of the Return to Education, 1995-2010 (with confidence intervals at 95%)
11
See Card (1998).
12 See Imbens and Angrist (1994), Angrist, Imbens and Rubin (1996).
4.00%
5.00%
6.00%
7.00%
8.00%
9.00%
10.00%
1995 1998 2000 2002 2004 2006 2008 2010
20
6.1.2 Experience
Looking at the dynamics of experience (Table 9), we observe different pattern for
each year of the sample: from 1995 to 2008 the experience profile is a concave
function, more or less steeper, while in 2010 it is approximately a linear function.
Then, we can affirm that the experience profile is not linear function (except for
2010) and that the estimates are quite stable over the time period considered.
Table 9 - Estimates of the Experience Profile, 1995-2010
6.1.3 Control Variables
If we consider the DUMMY_MALE variable, we observe a strong evidence of a
gender pay gap, in favor of men for all the period considered, with an increasing
trend, passing from 13.2% in 1995 to 15.4% in 2010.
Looking at the geographical variables (DUMMY_NORTH and
DUMMY_SOUTH), we observe that when the estimates are significant, the
DUMMY_NORTH is positive while the other one in negative. This means that it is
more convenient to work in the north regions in comparison to the centre regions,
instead if an individual works in the south region he will earns less than in the
center regions. Then, working in the same sector of the father or the mother
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0 2 4 6 8 10 12 14 16 18 20 22 24 26 28 30 32 34 36 38 40
Years of experience
1995
1998
2000
2002
2004
2006
2008
2010
21
(DUMMY_SECT_PARENTS) seems to not bring particular benefits, except for
year 1998 and 2008 where this dummy is significant and positive.
Finally, considering different sector of employment, working in the agricultural
sector is less convenient than working in the industrial sector. On the contrary,
looking at the public sector, when the dummy is significant then it is positive,
meaning that working in the public sector is more convenient than working in the
industrial sector.
6.2 Estimation of the Returns for Different Type of School
The empirical specification in eq. (6) is based on the assumption that the return to
education is constant and independent of the level of attained education. We allow
the marginal return to schooling to vary with the level of completed education by
replacing years of schooling with three educational dummies, one for each level of
completed schooling above compulsory school, that is vocational school, secondary
and tertiary education.
This is the multiple factor model, an alternative way to estimate returns to
schooling, where different educational levels have separate effects on earnings.
As suggested by the 'credentialism' hypothesis, in the presence of heterogeneity
what really matters is the type of school rather than the overall number of years
spent in formal education. We investigate these issues by considering the highest
degree attained by individual using educational dummies rather than years of
schooling in our earnings regressions. In particular, we look at education
achievements by broad levels: compulsory school (no schooling, primary school
and junior high school), vocational, upper secondary and tertiary education, and
also we address the issue of "credentialism" by distinguishing among types of
school (for instance scientific, humanistic, technical) within upper secondary and
tertiary education.
Also in the case of the estimate the returns of education from different type of
school, we deal with the problem of endogeneity by using instrumental variables.
22
We apply the two step methodology proposed by Vella and Gregory (1996). The
empirical strategy consists of estimating the two following equations:
(9)
(10)
where is the real hourly wage, are educational dummies that correspond to the
highest degree achieved by the individual, and are vectors of observed
attributes, and are normally distributed error terms with zero means and finite
variances, is the latent level of education. We define as the observed level of
education, that takes the following discrete values:
(11)
and associate to the educational dummies by setting and
otherwise.
We use a two step procedure to estimate the coefficients. In the first step we
estimate an ordered probit model for educational attainment as a function of the
instrument used in the previous IV estimation. In the second step, we include the
score13
associated to the ordered probit in the earnings equation and then we apply
ordinary least squares. This method is closely related to instrumental variables
estimation.
Our specification of the ordered probit includes the same covariates of the
instrumental equation used before.
The interpretation of the estimated coefficients is in terms of additional return that
the educational level grants to the individual with respect to the reference group
that is compulsory school. Our results are reported in table 10. For instance, in
2010, an employee with a high school degree earns, on average, 36.8% more than
an employee with the same covariate belonging to the reference group. This
differential increase to 74.6% for graduated individuals.
13
See Idson and Feaster (1990) for details on the computation of the score.
23
The estimated coefficients of the score have always a negative sign when they are
significant, implying that the covariance between unobservable variables that affect
earnings and educational choice is negative. This means that an individual attains a
lower educational level than predicted, because abler individuals have a higher
marginal cost of schooling in terms of foregone earnings, due to more attractive
wage offer. Hence, these individuals tend to acquire less education that predicted
education and earn higher wages.
Table 10 – Second stage OLS estimates. Dependent Variable: log of hourly earnings less tax.
Omitted categories are: Center (DUMMY_CENTER); Industrial sector (DUMMY_INDUSTRIAL).
VARIABLES 1995 1998 2000 2002 2004 2006 2008 2010
VOCATIONAL 0.210*** 0.0631 0.214*** 0.218*** 0.138*** 0.158*** 0.0670* 0.150*** (0.0349) (0.0558) (0.0333) (0.0368) (0.0403) (0.0313) (0.0392) (0.0570)
UPPER_SECONDARY 0.372*** 0.253*** 0.379*** 0.354*** 0.223*** 0.351*** 0.278*** 0.368***
(0.0304) (0.0505) (0.0363) (0.0439) (0.0556) (0.0404) (0.0398) (0.0578) TERTIARY 0.752*** 0.513*** 0.768*** 0.739*** 0.637*** 0.740*** 0.606*** 0.746***
(0.0508) (0.0838) (0.0570) (0.0827) (0.0880) (0.0682) (0.0695) (0.105)
EXPERIENCE 0.0213*** 0.0191*** 0.0237*** 0.0281*** 0.0155*** 0.0264*** 0.0242*** 0.0160*** (0.00328) (0.00669) (0.00348) (0.00509) (0.00433) (0.00355) (0.00435) (0.00437)
EXPERIENCE^2 -0.0003*** -0.0002 -0.0003*** -0.0004*** -0.0002* -0.0004*** -0.0003*** -6.27e-05
(7.85e-05) (0.00015) (7.96e-05) (0.000122) (0.000111) (8.39e-05) (0.000111) (0.000101) DUMMY_MALE 0.135*** 0.124*** 0.100*** 0.0986*** 0.0683*** 0.109*** 0.156*** 0.160***
(0.0256) (0.0344) (0.0183) (0.0244) (0.0265) (0.0201) (0.0215) (0.0268)
DUMMY_MARRIED 0.00754 0.0535 0.0525** 0.00591 0.0434 -0.00679 -0.0446 0.0209 (0.0254) (0.0447) (0.0244) (0.0370) (0.0310) (0.0222) (0.0276) (0.0276)
NCOMP 0.0165** 0.0111 0.00168 0.00227 -0.00471 0.0322*** 0.0249*** -0.000861
(0.00734) (0.0144) (0.00745) (0.0101) (0.00910) (0.00840) (0.00860) (0.0108) DUMMY_HOUSEHOLD -0.00255 -0.00837 0.00911 0.0269 0.0267 0.0376** 0.0126 0
(0.0257) (0.0352) (0.0179) (0.0240) (0.0239) (0.0187) (0.0237) (0)
DUMMY_TOWN 0.00673 0.0374 0.00621 -0.0793** -0.00872 0.0443* 0.0191 -0.0296 (0.0211) (0.0411) (0.0209) (0.0349) (0.0422) (0.0239) (0.0304) (0.0373)
DUMMY_NORTH 0.0366** 0.0775*** 0.0467*** 0.0494** 0.0599** -0.00513 0.00316 0.0606**
(0.0168) (0.0282) (0.0162) (0.0232) (0.0282) (0.0182) (0.0227) (0.0283) DUMMY_SOUTH -0.0345* 0.0428 -0.0284 -0.0114 -0.00933 -0.0761*** -0.0489** 0.0167
(0.0187) (0.0314) (0.0214) (0.0273) (0.0330) (0.0219) (0.0249) (0.0289)
DUMMY_AGRICULTURAL -0.118 -0.143 -0.192*** -0.0991* -0.129*** -0.151** -0.0997** -0.00494 (0.0717) (0.0921) (0.0581) (0.0577) (0.0406) (0.0746) (0.0502) (0.0694)
DUMMY_PUBLIC 0.114*** 0.0991*** 0.0224 0.0214 0.0935*** 0.0465* 0.105*** 0.0545
(0.0227) (0.0330) (0.0229) (0.0312) (0.0346) (0.0282) (0.0273) (0.0390) DUMMY_OTHER_SECTOR 0.0103 0.0165 -0.000192 -0.00674 0.00655 -0.0132 0.00673 0.00186
(0.0179) (0.0387) (0.0198) (0.0232) (0.0260) (0.0194) (0.0235) (0.0263)
DUMMY_SECT_PARENTS 0.00569 0.0543** -0.0110 -0.000539 0.00188 -0.0120 0.0292 0.00830 (0.0180) (0.0255) (0.0146) (0.0185) (0.0188) (0.0160) (0.0189) (0.0215)
DUMMY_PART_TIME 0.0355 0.0618 0.0610* -0.0749* -0.0360 0.000124 0.0177 0.0393
(0.0346) (0.0633) (0.0330) (0.0430) (0.0391) (0.0398) (0.0355) (0.0368) SCORE -0.0543*** 0.00663 -0.0953*** -0.0772*** -0.0362 -0.0856*** -0.0436* -0.0960***
(0.0187) (0.0317) (0.0215) (0.0275) (0.0319) (0.0238) (0.0255) (0.0369)
Constant 1.519*** 1.540*** 1.543*** 1.558*** 1.746*** 1.551*** 1.577*** 1.544*** (0.0462) (0.0985) (0.0459) (0.0612) (0.0568) (0.0597) (0.0545) (0.0705)
Observations 4,352 1,468 3,783 3,321 3,405 3,437 2,836 2,145 R-squared 0.412 0.317 0.338 0.287 0.245 0.327 0.345 0.295
Robust standard errors in parentheses
*** p<0.01, ** p<0.05, * p<0.1
Looking at the different level of education, vocational school seems to have a not
clear pattern, passing from 21% in 1995 to 15% in 2010.
24
The rate of return of secondary school is not constant over the period considered,
but it shows a slightly decreasing trend from 1995 to 2010. The same trend is
observed for the rate of return of tertiary education (university).
Table 11 – Rate of Return of Different Type of School 1995-2010
However, even if the college premium does not have a particular trend, going to
college allows to have between 30% and 40% of high wages (excluding year 1998).
Table 12 – Annual Rate of Return of Different Type of School 1995-2010
(reference category: compulsory school)
Moreover, we assume that these returns can be spread evenly among the years of
school required to complete a degree (see Table 12). It turns out that the increase in
0.00%
5.00%
10.00%
15.00%
20.00%
25.00%
30.00%
35.00%
40.00%
45.00%
1995 1998 2000 2002 2004 2006 2008 2010
Vocational
Upper Secondary
Tertiary
0.00%
1.00%
2.00%
3.00%
4.00%
5.00%
6.00%
7.00%
8.00%
1995 1998 2000 2002 2004 2006 2008 2010
Vocational
Upper Secondary
Tertiary
25
earnings due to an additional year of vocational school, upper secondary school and
college is respectively 5%, 7.4% and 7.6% for 2010. Hence, there is evidence that
returns to education are not constant but increase with the level of attained
education.
Finally, considering experience and the other control variables that are included in
the estimation, we do not observe significant changes from the IV estimates.
7 Concluding Remarks
We have studied the economic returns to education in Italy using cross-sectional
data from the 1995 to 2010 waves of the Bank of Italy survey on the income and
wealth of Italian household. We apply instrumental variables estimation to solve
the problem of endogeneity. The evidence is that returns to schooling have not
changed much over the period considered, 1995-2010, and are between 5.9% and
7.9%, recording the highest level for 2006. Looking to the different sector of
employment, a relative convenience to work in the public sector emerges. In
addition, there is an evidence of a gender pay gap, in favor of men for all the period
considered.
When the type of school attended is taken into consideration, we also find that the
returns to education increase with higher levels of educational attainment. In this
case, to solve the problem of endogeneity, an ordered probit is applied to the choice
of educational attainment and then we add the score of the probit estimation, to the
original equation and apply OLS. In particular, for 2010, the estimated coefficient
of the educational dummy is respectively 15% for vocational school, 36.8% for
upper secondary, and 74.6% college education. Abler individuals, who received
better wage offers, have lower education than predicted, because of the relative
incentive to anticipate labor market entry (as signaled by the negative coefficient of
the score).
In this analysis we take into consideration only employees excluding self-employed
because of low reliability of their declared earnings. Restricting the analysis only to
employees probably leads to an underestimation of the returns to education in Italy.
26
However, the possible presence of outliers in earnings of certain categories of self-
employed (typically professionals and managers) could lead to an upward bias. The
use of quantile regression could be the solution to this problem and is left to future
research.
27
References
Acemoglu, D. and J. D. Angrist (2001). How Large are Human-Capital
Externalities? Evidence from Compulsory-Schooling Laws. In B. Bernanke and
K. Rogoff, NBER Macroeconomics Annual (9-74). Cambridge: National Bureau
of Economic Research.
Angrist, J. D., G. W. Imbens, D. B. Rubin (1996). Identification of Causal Effects
Using Instrumental Variables, Journal of the American Statistical Association,
V. 91, 444- 455.
Angrist, J. D. and Newey, W. (1991). Over-Identification Test in Earnings Function
with Fixed Effects, Journal of Business and Economic Statistics, July, 317-323.
Angrist, J. D. and A. B. Krueger, (1991). Does Compulsory Schooling Affect
Schooling And Earnings?, Quarterly Journal of Economics, November, 979-
1014.
Angrist, J. D. and A. B. Krueger, (1992). Estimating the Payoff to Schooling Using
the Vietnam Era Draft Lottery, NBER Working Paper No. 4067.
Antonelli, G. (1985). Risorse Umane e Redditi da lavoro. Franco Angeli, Milan.
Ashenfelter, O. and A. B. Krueger (1994). Estimates of the Economic Return to
Schooling from a New Sample of Twins”, American Economic Review,
December, 84(5), 1157-73.
Ashenfelter, O. and D. J. Zimmerman (1993). Estimates of the Return to Schooling
from Sibling Data: Fathers, Sons, and Brothers”, NBER Working Paper No.
4491.
Ashenfelter, O., Harmon, C and H. Oosterbeek (1999). A Review of the Estimates
of the Schooling/Earnings Relationship, with Tests for Pubblication Bias, Labour
of Economis, 6 (3), 453-470.
Becker, G. S. (1962). Investment in Human Capital: A theoretical Analysis, The
Journal of Political Economy, 70 (5), 9-49.
Becker, G. S. (1991). Treatise on the Family, Cambridge: Harvard University
Press. (first ed. 1981)
Becker, G. S. (1993). Human Capital: A Theoretical and Empirical Analysis with
Special Reference to Education, Chicago: University of Chicago Press (first ed.
1975).
28
Ben-Porath, Y. (1967). The Production of Human Capital and the Life Cycle of
Earnings, Journal of Political Economy, 75(4), part 1, 352-65.
Bound, J., Jaeger, D. and R. Baker (1995). Problems with instrumental variables
estimation when the correlation between the instruments and the endogenous
explanatory variables is weak, Journal of the American Statistical Association,
443-450.
Bowman, A.W. and A. Azzalini (1997). Applied Smoothing Techniques for Data
Analysis: the Kernel Approach with S-Plus Illustrations. Oxford University
Press, Oxford.
Brandolini A., Cannari L. (1994). “Methodological appendix: the Bank of Italy's
Survey of Household income and wealth”, in Ando A., Guiso L., Visco I. (a cura
di), Savings and the accumulation of wealth. Essays on Italian households and
government saving behaviour, Cambridge University Press.
Brunello, G., Comi, S. and C. Lucifera (1999). The Returns to Education in Italy: A
Review of the Applied Literature, in R. Asplund and P.T. Pereira, Returns to
Human Capital in Europe. A Literature Review, ETLA, Helsinki.
Brunello, G., Comi, S. and C. Lucifera (2000). The Returns to Education in Italy: A
new look at evidence, IZA DP No. 130.
Brunello, G. and R. Miniaci (1999). The Economic Returns to Schooling for Italian
men. An Evaluation Based on Instrumental Variables, Labour Economics 6,
509–519.
Cannari, L., and G. D’Alessio (1995). Il Rendimento Dell’Istruzione: Alcuni
Problemi Di Stima. Temi di Discussione n. 253, Bank of Italy, Rome.
Cannari, L., Pellegrini, G. and P. Sestito (1989). Età, Istruzione e Retribuzioni,
Contributi all’analisi economica, Bank of Italy, Rome.
Card, D. (1994). Earnings, Schooling and Ability Revisited. In S. Polacheck (ed.),
Research in Labor Economics, Vol. 14, Greenwich: JAI Press. (NBER, Working
Paper No. 4832)
Card, D. (1999). The Causal Effect of Education on Earnings. In Ashenfelter, O.
and D. Card, Handbook of Labor Economics, Rotterdam: Elsevier.
Card, D. (2001). Estimating the Return to Schooling: Progress on Some Persistent
Econometric Problems, Econometrica, 69 (5), 1127-60.
Chiswick, B. (2003). Jacob Mincer, Experience and the Distribution of Earnings,
Review of Economics of the Household 1, 343-361.
29
Ciccone, A. (2004). Human Capital as a Factors of Growth and Employment at the
Regional Level. The Case of Italy. Report for the European Commission, DG for
Employment and Social Affairs.
Ciccone, A., Cingano, F. and P. Cipollone, (2006). The private and social return to
schooling in Italy, Temi di Discussione n. 569, Bank of Italy, Rome.
Cingano, F., and P. Cipollone (2009). I rendimenti dell’istruzione, Questioni di
Economia e Finanza (Occasional Papers) n.53, Bank of Italy, Rome.
Colussi, A. (1997). Il Tasso Di Rendimento dell’Istruzione In Italia: un’Analisi
Cross-Section, Politica Economica 13 (2), 273-294.
De La Fuente, A. and A. Ciccone (2002). Human Capital in a Global and
Knowledge-Based Economy, European Commission: DG Employment and
Social Affairs.
Flabbi, L. (1997). Investire In Istruzione: Meglio per lui o per lei?, mimeo, Bocconi
University, Milan.
Flabbi, L. (1999). Returns to Schooling in Italy: OLS, OV and Gender Differences.
Working paper of Bocconi University, Serie di econometria ed economia
applicata.
Garen, J. (1984). The returns to schooling: a selectivity bias approach with a
continuous choice variable. Econometrica, 52, 1199-1218.
Griliches, Z. (1977). Estimating the Returns to Schooling: Some Econometric
Problems, Econometrica, 45, 1, 1-22.
Harmon, C.P. and I. Walker (1995). Estimates of the Economic Return to
Schooling for the UK, American Economic review, 1278-86.
Harmon, C.P., Oosterbeek, H and I. Walker (2003). The Returns to Education:
Microeconometrics, Journal of Economic Surveys, 17(2), 115-155.
Heckman, J. J. (1979). Sample Selection Bias as A Specification Error,
Econometrica, 47(1), 153-161.
Heckman, J. J. and M.R. Killingsworth (1986). Female Labor Supply: a Survey, in
Ashenfelter O. and R. Layard (ed.) Handbook of Labor Economics, Amsterdam:
North Holland.
Heckman, J. J., Lochner, L. and P. Todd (2003). Fifty Years of Mincer Earnings
Regressions, NBER, Working Paper no. 9732.
30
Heckman, J. J., Lochner, L. and P. Todd (2006). Earnings Functions, Rates of
Return and Treatment Effects: The Mincer Equation and Beyond, In E.
Hanushek and F. Welch, Handbook of the Economics of Education, 307-458,
Rotterdam: Elsevier.
Heckman, J. J., Lochner, L. and P. Todd (2008). Earnings Functions and Rates of
Return. Journal of Human Capital, 2 (1), 1-31.
Ichino, A. and R. Winter-Ebmer, (1999). Lower and Upper Bounds of Returns to
Schooling: An Exercise in IV Estimation with Different Instruments, European
Economic Review, 43, 889-901.
Idson, T. L. and D. J. Feaster (1991). A Selectivity Model of Employer-Size Wage
Differentials, Journal of Labor Economics, Vol. 8 (1), 99-122.
Imbens, G., Angrist, J.D., 1994. Identification and estimation of local average
treatment effects, Econometrica 62, 467-475.
Lam, D. and R.F. Schoeni (1993). Effects of Family Background on Earnings and
Returns to Schooling: Evidence from Brazil, Journal of Political Economy, Vol.
101 (4), 710-740.
Lucifora, C. and B. Reilly (1990). Wage Discrimination and Female Occupational
Intensity, Labour, n.4, 147-168.
Meghir, C. and S. G. Rivkin (2010). Econometric Methods for Research in
Education, NBER, Working Paper no. 16003.
Mendolicchio, C. (2006). A Disaggregate Analysis of Private Returns to Education
in Italy. Discussion Paper 54, Département des Sciences Économiques de
l'Université catholique de Louvain.
Mincer, J. (1974). Schooling, Experience and Earnings. New York: Columbia
University Press.
OECD (2011) Education at a Glance, Paris.
Peracchi, F. (2004). “Educational wage premia and the distribution of earnings: an
international perspectives”, in E.Hanushek and F.Welch (eds.), Handbook of the
Economics of Education, North-Holland, Amsterdam.
Psacharopoulos, G. (1994). Returns to Investment in Education: A Global Update,
World Development, v.22, n.9, 1325-1343.
Schultz, T. W. (1961). Investment in “Human Capital”, The American Economic
Review, 51 (1).
31
Sianesi, B. and J. V. Reenen (2003). The Returns to Education: Macroeconomics,
Journal of Economic Surveys, Vol. 17 (2), 157-200.
Spence, M. (1973). Job Market Signaling. Quarterly Journal of Economics, 87 (3),
355-373.
Vella, F. and R. G. Gregory (1996), Selection Bias and Human Capital Investment:
Estimating the Rates of Return to Eduation for Young Males, Labour
Economics: An International Journal, 197-219.
Vieira, J.A.C. (1999). Returns to Education in Portugal, Labour Economics, Vol. 6
(4), 535-541.
Willis, R. (1986). Wage Determinants: A Survey and Reinterpretation of Human
Capital Earnings Functions. In O. Ashenfelter and R. Layard (eds), Handbook of
Labor Economics, Amsterdam and New York: North Holland.
Wooldridge, J.M. (2002). Econometric Analysis of Cross Section and Panel Data.
MIT press.
32
Appendix
Table 13 – Mean of the Log of hourly earnings less tax (1995 -2010)
Table 14 – Mean of the number of year of Schooling (1995 -2010)
Table 15 – Mean of the number of year of Experience (1995 -2010)
2.12
2.14
2.16
2.18
2.2
2.22
2.24
2.26
1995 1998 2000 2002 2004 2006 2008 2010
10.0
10.2
10.4
10.6
10.8
11.0
11.2
11.4
11.6
11.8
12.0
1995 1998 2000 2002 2004 2006 2008 2010
15
15.5
16
16.5
17
17.5
18
18.5
19
1995 1998 2000 2002 2004 2006 2008 2010
33
Table 16 shows OLS estimates, obtained by including in the original specification
controls for the composition of her/his family, the geographical area of residence
and the sector in which the individual is currently working.
The evidence is that returns to education are fairly stable from 1998 to 2010, but if
we compare the year 1995 with 2010, returns to education have decreased from
5.14% to 4.16%.
Moreover, Table 16 shows that returns are higher for male over the entire sample
period, confirming the gender pay gap.
Table 16 - OLS estimates. Dependent Variable: log of hourly earnings less tax. Omitted categories
are: Center (DUMMY_CENTER); Industrial sector (DUMMY_INDUSTRIAL).
VARIABLES 1995 1998 2000 2002 2004 2006 2008 2010
SCHOOL 0.0514*** 0.0447*** 0.0425*** 0.0454*** 0.0409*** 0.0450*** 0.0442*** 0.0416***
(0.00173) (0.00315) (0.00188) (0.00229) (0.00224) (0.00196) (0.00229) (0.00214)
EXPERIENCE 0.0272*** 0.0275*** 0.0254*** 0.0271*** 0.0210*** 0.0250*** 0.0275*** 0.0194***
(0.00277) (0.00458) (0.00246) (0.00285) (0.00300) (0.00268) (0.00274) (0.00249)
EXPERIENCE^2 -0.000352*** -0.000351*** -0.000363*** -0.000362*** -0.000308*** -0.000386*** -0.000405*** -0.000207***
(6.82e-05) (0.000114) (6.05e-05) (7.46e-05) (8.14e-05) (6.75e-05) (7.15e-05) (6.20e-05)
DUMMY_MALE 0.0855*** 0.0422 0.0769*** 0.106*** 0.0790*** 0.0904*** 0.112*** 0.116***
(0.0155) (0.0282) (0.0137) (0.0163) (0.0174) (0.0159) (0.0144) (0.0143)
DUMMY_MARRIED 0.0739*** 0.0666** 0.105*** 0.0616*** 0.0702*** 0.0536*** 0.0310** 0.0685***
(0.0165) (0.0331) (0.0148) (0.0190) (0.0180) (0.0161) (0.0157) (0.0153)
NCOMP -0.00338 0.00475 -0.0104* -0.00938 -0.0122* 0.0140** 0.00779 -0.00191
(0.00535) (0.0105) (0.00551) (0.00653) (0.00665) (0.00655) (0.00573) (0.00609)
DUMMY_HOUSEHOLD 0.0436*** 0.0463 0.0326** 0.0382** 0.0385** 0.0651*** 0.0502*** 0.0169
(0.0167) (0.0285) (0.0134) (0.0159) (0.0165) (0.0147) (0.0135) (0.0128)
DUMMY_TOWN 0.0333* 0.0209 0.0458*** -0.0370 0.0233 0.0587*** 0.0281 -0.0120
(0.0175) (0.0340) (0.0178) (0.0278) (0.0313) (0.0215) (0.0243) (0.0240)
DUMMY_NORTH 0.0404*** 0.0778*** 0.0481*** 0.0399** 0.0473** -0.00849 -0.0296* 0.0441***
(0.0140) (0.0241) (0.0136) (0.0169) (0.0198) (0.0160) (0.0164) (0.0170)
DUMMY_SOUTH -0.0379** 0.0570** -0.0292 0.00365 -0.0232 -0.0821*** -0.0783*** -3.73e-05
(0.0172) (0.0288) (0.0189) (0.0232) (0.0241) (0.0188) (0.0186) (0.0189)
DUMMY_AGRICULTURAL -0.117* -0.0967 -0.132*** -0.0424 -0.0935*** -0.168*** -0.00792 -0.0596
(0.0679) (0.0705) (0.0439) (0.0568) (0.0329) (0.0480) (0.0419) (0.0388)
DUMMY_PUBLIC 0.174*** 0.109*** 0.108*** 0.110*** 0.141*** 0.126*** 0.143*** 0.148***
(0.0168) (0.0268) (0.0150) (0.0182) (0.0197) (0.0188) (0.0190) (0.0176)
DUMMY_OTHER_SECTOR 0.0109 -0.00141 0.0294* 0.0102 0.00797 0.000988 -0.00310 0.0161
(0.0149) (0.0301) (0.0152) (0.0177) (0.0189) (0.0161) (0.0160) (0.0146)
DUMMY_SECT_PARENTS 0.00296 0.0797*** 0.0175 0.0138 0.00658 -0.00163 0.0340*** 0.0114
(0.0151) (0.0228) (0.0121) (0.0157) (0.0146) (0.0135) (0.0131) (0.0133)
DUMMY_PART_TIME 0.0734** 0.0345 0.0375 -0.0842** -0.0479 -0.00797 0.0293 -0.00400
(0.0324) (0.0525) (0.0274) (0.0365) (0.0311) (0.0301) (0.0268) (0.0216)
Constant 1.116*** 1.136*** 1.255*** 1.224*** 1.381*** 1.285*** 1.250*** 1.304***
(0.0408) (0.0708) (0.0355) (0.0420) (0.0460) (0.0438) (0.0437) (0.0459)
Observations 6,066 2,016 5,724 5,461 5,425 5,378 5,409 5,161
R-squared 0.450 0.366 0.353 0.306 0.261 0.326 0.353 0.327
Robust standard errors in parentheses *** p<0.01, ** p<0.05, * p<0.1
34
Table 17 shows OLS estimates of the empirical specification, including interaction
of the variable schooling with experience and with gender.
Table 17 - OLS estimates with interactions. Dependent Variable: log of hourly earnings less tax.
Omitted categories are: Center (DUMMY_CENTER); Industrial sector (DUMMY_INDUSTRIAL).
VARIABLES 1995 1998 2000 2002 2004 2006 2008 2010
SCHOOL 0.0471*** 0.0445*** 0.0291*** 0.0388*** 0.0420*** 0.0267*** 0.0242*** 0.0330***
(0.00663) (0.0108) (0.00625) (0.00831) (0.00657) (0.00555) (0.00553) (0.00539) EXPERIENCE 0.0216*** 0.0455*** 0.00859 0.0158* 0.0199*** 0.00516 -0.000305 0.00707
(0.00810) (0.0138) (0.00760) (0.00923) (0.00772) (0.00846) (0.00879) (0.00714)
EXPERIENCE^2 -0.000294 -0.000970*** -6.90e-05 -0.000172 -0.000413** -0.000174 0.000133 -0.000119 (0.000196) (0.000344) (0.000183) (0.000229) (0.000199) (0.000219) (0.000228) (0.000177)
SCHOOL*EXPER 0.000492 -0.00170 0.00137** 0.000929 -6.37e-07 0.00155** 0.00230*** 0.000962
(0.000704) (0.00112) (0.000639) (0.000824) (0.000702) (0.000678) (0.000699) (0.000590) SCHOOL*EXPER^2 -4.17e-06 5.86e-05* -2.26e-05 -1.48e-05 1.22e-05 -1.33e-05 -4.40e-05** -5.12e-06
(1.78e-05) (3.04e-05) (1.60e-05) (2.13e-05) (1.85e-05) (1.81e-05) (1.82e-05) (1.51e-05)
DUMMY_MALE 0.129*** -0.0537 0.0986** 0.173*** 0.200*** 0.146*** 0.137*** 0.236*** (0.0432) (0.0821) (0.0436) (0.0567) (0.0505) (0.0440) (0.0484) (0.0474)
SCHOOL*MALE -0.00403 0.00848 -0.00207 -0.00590 -0.0106** -0.00498 -0.00206 -0.00994**
(0.00360) (0.00645) (0.00366) (0.00482) (0.00426) (0.00362) (0.00418) (0.00395) DUMMY_MARRIED 0.0732*** 0.0700** 0.105*** 0.0611*** 0.0703*** 0.0572*** 0.0284* 0.0659***
(0.0166) (0.0323) (0.0148) (0.0190) (0.0182) (0.0161) (0.0156) (0.0153)
NCOMP -0.00398 0.00307 -0.0113** -0.01000 -0.0130* 0.0133** 0.00832 -0.00198 (0.00537) (0.0105) (0.00546) (0.00646) (0.00664) (0.00653) (0.00571) (0.00606)
DUMMY_HOUSEHOLD 0.0441*** 0.0438 0.0334** 0.0393** 0.0391** 0.0624*** 0.0491*** 0.0172
(0.0168) (0.0281) (0.0134) (0.0160) (0.0165) (0.0146) (0.0135) (0.0127) DUMMY_TOWN 0.0317* 0.0142 0.0465*** -0.0389 0.0216 0.0585*** 0.0266 -0.0177
(0.0175) (0.0342) (0.0178) (0.0278) (0.0312) (0.0214) (0.0241) (0.0239)
DUMMY_NORTH 0.0399*** 0.0745*** 0.0468*** 0.0402** 0.0481** -0.0101 -0.0320* 0.0437*** (0.0140) (0.0238) (0.0136) (0.0169) (0.0198) (0.0159) (0.0164) (0.0168)
DUMMY_SOUTH -0.0416** 0.0563* -0.0322* 0.00139 -0.0274 -0.0864*** -0.0825*** -0.00393
(0.0172) (0.0287) (0.0189) (0.0232) (0.0237) (0.0188) (0.0185) (0.0186) DUMMY_AGRICULTURAL -0.116* -0.108 -0.129*** -0.0435 -0.0942*** -0.173*** -0.00882 -0.0569
(0.0686) (0.0679) (0.0441) (0.0567) (0.0331) (0.0482) (0.0421) (0.0384) DUMMY_PUBLIC 0.173*** 0.111*** 0.104*** 0.106*** 0.138*** 0.120*** 0.140*** 0.141***
(0.0169) (0.0268) (0.0150) (0.0181) (0.0197) (0.0187) (0.0191) (0.0174)
DUMMY_OTHER_SECTOR 0.00895 -0.00333 0.0280* 0.0107 0.0108 1.25e-05 -0.00294 0.0146 (0.0151) (0.0301) (0.0150) (0.0179) (0.0190) (0.0161) (0.0158) (0.0145)
DUMMY_SECT_PARENTS 0.00368 0.0778*** 0.0178 0.0145 0.00647 0.00102 0.0353*** 0.0122
(0.0151) (0.0225) (0.0121) (0.0157) (0.0146) (0.0135) (0.0132) (0.0131) DUMMY_PART_TIME 0.0763** 0.0278 0.0397 -0.0803** -0.0440 -0.00795 0.0287 -0.00227
(0.0325) (0.0518) (0.0271) (0.0367) (0.0312) (0.0300) (0.0268) (0.0215)
Constant 1.168*** 1.153*** 1.424*** 1.307*** 1.375*** 1.518*** 1.496*** 1.416*** (0.0821) (0.144) (0.0766) (0.0980) (0.0819) (0.0812) (0.0766) (0.0773)
Observations 6,066 2,016 5,724 5,461 5,425 5,378 5,409 5,161 R-squared 0.451 0.371 0.356 0.308 0.264 0.334 0.358 0.333
35
Table 18 shows the estimates of the first stage regression of the instrumental
variables estimation.
Table 18 – First stage of IV estimates. Dependent Variable: schooling. Omitted categories are:
Center (DUMMY_CENTER); Industrial sector (DUMMY_INDUSTRIAL).
VARIABLES 1995 1998 2000 2002 2004 2006 2008 2010
SCHOOL_F 0.296*** 0.292*** 0.298*** 0.267*** 0.237*** 0.261*** 0.261*** 0.281*** (0.0213) (0.0393) (0.0248) (0.0253) (0.0240) (0.0247) (0.0289) (0.0431)
SCHOOL_M 0.216*** 0.207*** 0.150*** 0.172*** 0.189*** 0.135*** 0.183*** 0.145***
(0.0254) (0.0513) (0.0284) (0.0290) (0.0261) (0.0285) (0.0332) (0.0433) EXPERIENCE 0.0363 -0.0375 -0.0129 -0.00837 -0.0495 -0.0452 -0.0311 -0.0588
(0.0268) (0.0492) (0.0304) (0.0350) (0.0302) (0.0297) (0.0311) (0.0449)
EXPERIENCE^2 -0.00226*** 0.000503 -0.000842 -0.000697 0.000760 -0.000107 -0.000191 0.000484 (0.000662) (0.00118) (0.000739) (0.000844) (0.000727) (0.000696) (0.000719) (0.00101)
DUMMY_MALE 0.269 0.408 -0.332* -0.414** -0.395** -0.311 -0.332 -0.592***
(0.195) (0.340) (0.172) (0.190) (0.174) (0.193) (0.202) (0.219) DUMMY_MARRIED 0.510** -0.428 -0.0309 0.293 0.617*** 0.417** 0.792*** -0.0912
(0.246) (0.419) (0.243) (0.250) (0.221) (0.209) (0.251) (0.250)
NCOMP -0.169*** -0.196* -0.0198 0.0849 -0.0318 -0.0222 -0.170** 0.226** (0.0643) (0.114) (0.0754) (0.0777) (0.0754) (0.0732) (0.0784) (0.0980)
DUMMY_HOUSEHOLD -0.168 -0.629* 0.194 -0.0168 0.109 0.178 0.0705 0
(0.199) (0.344) (0.165) (0.181) (0.171) (0.180) (0.218) (0) DUMMY_TOWN 0.223 -0.179 0.903*** 0.529** 0.110 0.398* 0.347 0.435
(0.189) (0.354) (0.209) (0.247) (0.216) (0.225) (0.273) (0.292)
DUMMY_NORTH -0.158 0.0633 0.237 0.379** 0.0236 -0.233 0.173 -0.252 (0.159) (0.268) (0.168) (0.181) (0.172) (0.179) (0.214) (0.247)
DUMMY_SOUTH -0.197 0.0838 0.0493 0.310 -0.315 -0.621*** -0.241 -0.281
(0.177) (0.293) (0.190) (0.225) (0.222) (0.212) (0.240) (0.304) DUMMY_AGRICULTURAL -1.372*** -2.770*** -1.450*** -1.442*** -1.328*** -0.756** -1.059*** -1.098*
(0.466) (0.562) (0.371) (0.332) (0.335) (0.312) (0.397) (0.636)
DUMMY_PUBLIC 2.465*** 2.268*** 2.546*** 2.452*** 2.520*** 2.490*** 2.424*** 2.235*** (0.177) (0.285) (0.175) (0.205) (0.192) (0.175) (0.201) (0.294)
DUMMY_OTHER_SECTOR 0.00794 0.473 0.806*** 0.502*** 0.865*** 0.715*** 0.646*** 0.643***
(0.159) (0.300) (0.172) (0.176) (0.180) (0.176) (0.192) (0.229) DUMMY_SECT_PARENTS 0.154 0.273 0.290** -0.161 0.182 0.393*** 0.192 -0.0290
(0.151) (0.218) (0.135) (0.151) (0.139) (0.144) (0.163) (0.204)
DUMMY_PART_TIME -0.629*** -0.659* -0.681*** -0.553** -0.819*** -0.856*** -0.601** -0.822** (0.228) (0.377) (0.226) (0.261) (0.248) (0.237) (0.292) (0.411)
Constant 7.434*** 8.903*** 7.910*** 7.579*** 7.968*** 8.905*** 8.515*** 9.259***
(0.385) (0.732) (0.398) (0.467) (0.412) (0.423) (0.461) (0.599)
Observations 4,352 1,468 3,783 3,321 3,405 3,437 2,836 2,145
R-squared 0.408 0.408 0.391 0.373 0.365 0.365 0.364 0.322 Sargan test χ2(1) 1.691 1.891 0.239 0.05 0.515 0.197 0.026 0.457
p-Value 0.1935 0.1691 0.6248 0.8239 0.473 0.657 0.8716 0.5038
F-test on excl. instrum. 461.28 147.101 288.83 201.88 225.05 195.76 188.55 107,67
p-Value 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000
Robust standard errors in parentheses *** p<0.01, ** p<0.05, * p<0.1
Table 19 reports the results of both OLS estimates that use educational dummies
rather than years of schooling. The estimated coefficient of educational dummies
should be interpreted in terms of the additional return that the combination of
educational level plus school type grants to the individual with respect to the
reference category (compulsory school14
).
14
We include no schooling, primary school and junior high school.
36
Table 19 - OLS estimates. Dependent Variable: log of hourly earnings less tax. Omitted categories
are: compulsory school (no schooling, primary school and junior high school); diploma
professionale (DIP_PROF); laurea scientific (L_SCIEN); Center (DUMMY_CENTER); Industrial
sector (DUMMY_INDUSTRIAL).
VARIABLES 1995 1998 2000 2002 2004 2006 2008 2010
VOCATIONAL 0.162*** 0.0949** 0.0786*** 0.0962*** 0.0777*** 0.0981*** 0.0278 0.0452**
(0.0258) (0.0388) (0.0191) (0.0234) (0.0221) (0.0220) (0.0204) (0.0228) UPPER_SECONDARY 0.230*** 0.151*** 0.153*** 0.162*** 0.0856** 0.104*** 0.154*** 0.148***
(0.0261) (0.0342) (0.0269) (0.0299) (0.0371) (0.0241) (0.0266) (0.0253)
DIPL_TECN 0.0504* 0.0581 0.0627** 0.0430 0.0548 0.114*** 0.0277 0.0328 (0.0276) (0.0357) (0.0269) (0.0305) (0.0396) (0.0257) (0.0278) (0.0266)
DIPL_LIC 0.0896*** 0.0794 0.0772** 0.0797** 0.136*** 0.0882*** 0.0588 0.0236
(0.0347) (0.0586) (0.0376) (0.0387) (0.0463) (0.0321) (0.0362) (0.0340) DIPL_MAG 0.145*** 0.240*** 0.103*** 0.127** 0.134*** 0.227*** 0.152*** 0.0986***
(0.0336) (0.0740) (0.0360) (0.0515) (0.0475) (0.0358) (0.0362) (0.0345)
DIP_OTHER -0.0391 0.188** 0.119** -0.0293 0.0834 0.0351 0.0532 0.178* (0.0608) (0.0778) (0.0493) (0.0666) (0.0594) (0.0607) (0.0488) (0.0968)
TERTIARY 0.621*** 0.568*** 0.506*** 0.582*** 0.561*** 0.497*** 0.498*** 0.522***
(0.0341) (0.0482) (0.0322) (0.0426) (0.0415) (0.0370) (0.0386) (0.0331) L_LETT 0.0281 -0.0776 0.0146 -0.121** -0.00562 0.0203 0.0310 -0.0515
(0.0481) (0.0671) (0.0493) (0.0558) (0.0615) (0.0500) (0.0660) (0.0436)
L_UMAN -0.0577 -0.0562 -0.0325 -0.0120 -0.140*** -0.0474 -0.0575 -0.0808* (0.0545) (0.0676) (0.0461) (0.0676) (0.0528) (0.0486) (0.0451) (0.0480)
L_OTHER -0.0526 -0.138 -0.128* -0.242** -0.0682 -0.0366 -0.0490 -0.204*** (0.0695) (0.0966) (0.0662) (0.115) (0.0636) (0.0604) (0.0864) (0.0561)
EXPERIENCE 0.0284*** 0.0285*** 0.0274*** 0.0286*** 0.0228*** 0.0259*** 0.0289*** 0.0204***
(0.00274) (0.00461) (0.00246) (0.00281) (0.00296) (0.00265) (0.00278) (0.00253) EXPERIENCE^2 -0.0004*** -0.0004*** -0.0004*** -0.0004*** -0.00037*** -0.0004*** -0.00045*** -0.0002***
(6.73e-05) (0.000115) (6.06e-05) (7.35e-05) (7.97e-05) (6.70e-05) (7.27e-05) (6.32e-05)
DUMMY_MALE 0.102*** 0.0660** 0.0852*** 0.109*** 0.0872*** 0.102*** 0.127*** 0.122*** (0.0161) (0.0274) (0.0142) (0.0162) (0.0176) (0.0157) (0.0144) (0.0144)
DUMMY_MARRIED 0.0664*** 0.0625* 0.0962*** 0.0532*** 0.0610*** 0.0485*** 0.0238 0.0607***
(0.0161) (0.0328) (0.0149) (0.0192) (0.0175) (0.0163) (0.0153) (0.0151) NCOMP -0.00215 0.00744 -0.00791 -0.00793 -0.0108 0.0170*** 0.00881 -0.000711
(0.00530) (0.0103) (0.00542) (0.00642) (0.00663) (0.00646) (0.00567) (0.00592)
DUMMY_HOUSEHOLD 0.0423** 0.0517* 0.0319** 0.0400** 0.0396** 0.0688*** 0.0506*** 0.0169 (0.0166) (0.0286) (0.0133) (0.0159) (0.0162) (0.0147) (0.0133) (0.0126)
DUMMY_TOWN 0.0362** 0.0167 0.0357** -0.0415 0.0199 0.0588*** 0.0198 -0.0135
(0.0178) (0.0333) (0.0177) (0.0270) (0.0302) (0.0216) (0.0240) (0.0235) DUMMY_NORTH 0.0380*** 0.0837*** 0.0497*** 0.0424** 0.0403** -0.00494 -0.0238 0.0482***
(0.0141) (0.0246) (0.0135) (0.0171) (0.0191) (0.0161) (0.0164) (0.0172)
DUMMY_SOUTH -0.0495*** 0.0435 -0.0432** -0.00330 -0.0404* -0.0952*** -0.0924*** -0.00475 (0.0173) (0.0288) (0.0186) (0.0231) (0.0232) (0.0185) (0.0184) (0.0188)
DUMMY_AGRICULTURAL -0.172** -0.162** -0.174*** -0.0812 -0.129*** -0.175*** -0.0275 -0.0829**
(0.0674) (0.0651) (0.0426) (0.0574) (0.0318) (0.0469) (0.0427) (0.0385) DUMMY_PUBLIC 0.162*** 0.0973*** 0.105*** 0.105*** 0.128*** 0.121*** 0.134*** 0.134***
(0.0169) (0.0255) (0.0150) (0.0186) (0.0197) (0.0189) (0.0188) (0.0172)
DUMMY_OTHER_SECTOR 0.00640 0.00866 0.0360** 0.0146 0.0158 0.00982 0.00909 0.0168 (0.0149) (0.0303) (0.0151) (0.0180) (0.0186) (0.0163) (0.0156) (0.0144)
DUMMY_SECT_PARENTS 0.0120 0.0741*** 0.0179 0.0135 0.0101 -0.00224 0.0331** 0.0128
(0.0149) (0.0220) (0.0119) (0.0155) (0.0142) (0.0133) (0.0130) (0.0131) DUMMY_PART_TIME 0.0680** 0.0397 0.0336 -0.0823** -0.0513* -0.00954 0.0300 -0.00800
(0.0323) (0.0533) (0.0267) (0.0360) (0.0311) (0.0300) (0.0270) (0.0214)
Constant 1.488*** 1.459*** 1.570*** 1.575*** 1.707*** 1.628*** 1.596*** 1.637*** (0.0350) (0.0605) (0.0303) (0.0359) (0.0376) (0.0368) (0.0335) (0.0359)
Observations 6,066 2,016 5,724 5,461 5,425 5,378 5,409 5,161 R-squared 0.454 0.379 0.359 0.313 0.281 0.332 0.364 0.338
Robust standard errors in parentheses
*** p<0.01, ** p<0.05, * p<0.1
For year 2010, we find that, conditional on experience and other characteristics,
individuals who have completed upper secondary school earn 14.8% more than
individuals with only compulsory education. The earnings of individuals with
upper secondary who have completed college are 52.2%.
37
We confirm that the estimated returns to education are not constant but increase
with the level of education attained. Then, it confirms the positive monotonic
relationship that links returns to education to the highest level of education attained.
Table 20 reports the results of the ordered probit model for educational attainment
as a function of the instrument used in the IV estimation. This is the first step
necessary to estimate the score associated to the ordered probit that we add in the
earnings equation in order to apply ordinary least squares as second step.
Table 20 – Ordered probit estimates. Dependent Variable: education. Omitted categories are:
Center (DUMMY_CENTER); Industrial sector (DUMMY_INDUSTRIAL).
VARIABLES 1995 1998 2000 2002 2004 2006 2008 2010
PRIMARY_F -0.836*** -0.716*** -0.586*** -0.564***
(0.0653) (0.0762) (0.0774) (0.0912) PRIMARY_M -0.495*** -0.293***
(0.0729) (0.0775)
EXPERIENCE 0.00682 -0.0206 -0.0173 -0.0142 -0.0385*** -0.0194 -0.0130 -0.0178 (0.0102) (0.0189) (0.0111) (0.0133) (0.0121) (0.0125) (0.0127) (0.0157)
EXPERIENCE^2 -0.000571** 0.000451 0.000125 8.52e-05 0.000781*** 3.78e-05 3.49e-05 0.000118
(0.000254) (0.000449) (0.000272) (0.000319) (0.000283) (0.000301) (0.000294) (0.000364) DUMMY_MALE 0.0436 0.112 -0.185*** -0.149** -0.174** -0.115 -0.137* -0.240***
(0.0738) (0.140) (0.0648) (0.0717) (0.0687) (0.0810) (0.0766) (0.0805)
DUMMY_MARRIED 0.194** -0.188 0.0617 0.112 0.292*** 0.130 0.342*** -0.0254 (0.0983) (0.159) (0.0881) (0.0984) (0.0896) (0.0869) (0.0979) (0.0946)
NCOMP -0.0637** -0.117*** -0.00799 0.0140 -0.0326 -0.0215 -0.0469 0.0634*
(0.0247) (0.0443) (0.0263) (0.0304) (0.0302) (0.0308) (0.0312) (0.0364)
DUMMY_HOUSEHOLD -0.0893 -0.321** 0.0665 -0.0681 0.0224 0.0294 0.0408
(0.0749) (0.140) (0.0624) (0.0698) (0.0681) (0.0792) (0.0856)
DUMMY_TOWN 0.113 0.0409 0.295*** 0.193** 0.0735 0.220** 0.149 0.190* (0.0703) (0.126) (0.0783) (0.0957) (0.0901) (0.0885) (0.102) (0.106)
DUMMY_NORTH -0.0315 -0.0376 0.142** 0.148** 0.0359 -0.130* 0.0135 -0.103
(0.0614) (0.100) (0.0643) (0.0729) (0.0714) (0.0724) (0.0830) (0.0889) DUMMY_SOUTH -0.0787 0.0320 0.0551 0.0977 -0.134 -0.291*** -0.196** -0.192*
(0.0675) (0.106) (0.0704) (0.0876) (0.0868) (0.0877) (0.0941) (0.106)
DUMMY_AGRICULTURAL -0.426** -1.421*** -0.397*** -0.564*** -0.770*** -0.528*** -0.463** -0.334 (0.191) (0.282) (0.147) (0.199) (0.183) (0.167) (0.186) (0.251)
DUMMY_PUBLIC 0.929*** 0.749*** 0.947*** 0.976*** 0.987*** 0.991*** 0.922*** 0.892***
(0.0682) (0.113) (0.0646) (0.0760) (0.0760) (0.0736) (0.0790) (0.104) DUMMY_OTHER_SECT 0.0287 0.0761 0.300*** 0.256*** 0.342*** 0.309*** 0.285*** 0.298***
(0.0651) (0.116) (0.0670) (0.0719) (0.0735) (0.0709) (0.0775) (0.0869)
DUMMY_SECT_PARENTS 0.0724 0.127 0.135*** -0.00387 0.117** 0.221*** 0.110* 0.0442 (0.0616) (0.0843) (0.0505) (0.0592) (0.0560) (0.0585) (0.0629) (0.0734)
DUMMY_PART_TIME -0.264** -0.338** -0.218** -0.164 -0.233** -0.288*** -0.193* -0.327**
(0.104) (0.156) (0.0939) (0.108) (0.0969) (0.0959) (0.110) (0.141) SECONDARY_F 0.950*** 0.628*** 0.676*** 0.726***
(0.109) (0.0700) (0.0735) (0.0807)
SECONDARY_M 0.376*** 0.422*** 0.553*** 0.436*** 0.453*** 0.280*** (0.120) (0.0833) (0.0838) (0.0784) (0.0849) (0.0989)
cut1
Constant -1.009*** -0.484* -0.416*** -0.247 0.148 -0.399** 0.194 -0.799***
(0.153) (0.256) (0.157) (0.184) (0.165) (0.174) (0.181) (0.220)
cut2
Constant -0.771*** -0.226 -0.105 0.0291 0.421** -0.0802 0.515*** -0.424* (0.153) (0.254) (0.157) (0.184) (0.165) (0.176) (0.183) (0.220)
cut3
Constant 0.649*** 1.160*** 1.233*** 1.407*** 1.881*** 1.379*** 1.927*** 0.855***
(0.153) (0.258) (0.158) (0.188) (0.173) (0.181) (0.194) (0.221)
Observations 4,352 1,468 3,783 3,321 3,405 3,437 2,836 2,145
Robust standard errors in parentheses
*** p<0.01, ** p<0.05, * p<0.1